[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

Backgammon supercomputing

Posted By: Frank Berger
Date: Saturday, 17 May 2008, at 9:42 a.m.

In Response To: Backgammon supercomputing (Maik Stiebler)

Lastly, I wouldn't have thought that 32-bit precision is a problem. Could you expand a bit on why you think it is?

Think about matches. When you use have long matches and e.g. a nearly decided running game the difference between the moves is sometimes even with 64-Bit so small numerical artefacts show up (i.e. nuts moves). Therefore I don't use MET for move ranking unless both players have less than 50 points to go.

With bgb2go I used 32-Bit multiplication and a lot of nuts moved showed up, especially in nearly decided situations. Therefore we now use 64-Integer multiplication with 1.5 that was much better.

It's interesting that you find the Cell processors more promising. I seem to recall that in peak performance terms, the gpus offer much more bang for the buck. I assume the programming model for the Cells is much more flexible!?
I only invested a few hours each so I'm far from competent at that, but the programming modell looked more like a normal CPU to me.

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.